July 28: Researchers from the University of Wisconsin Madison and Argonne National Laboratory have developed a machine learning model that can track atmospheric conditions in near real time, improving predictions of wildfire smoke movement and helping authorities issue faster air quality warnings.
The study focuses on the atmospheric boundary layer the lowest part of the atmosphere where smoke, dust and other airborne pollutants mix. The height of this layer changes throughout the day, shrinking to less than 100 metres at night and expanding to around 2,000 metres during warmer daytime conditions. These changes determine whether wildfire smoke remains trapped near the ground or disperses into the atmosphere.
Using high resolution data from the U.S. Department of Energy CROCUS Urban Integrated Field Laboratory researchers developed a machine learning model that estimates boundary layer height in near real time, providing more accurate and timely information than conventional forecasting methods.
Traditional techniques rely on Doppler LiDAR, which measures atmospheric conditions using laser pulses. However, the data can be distorted by clouds, rainfall and even insect swarms, often requiring observations to be averaged over 30 minutes to an hour, making it difficult to detect rapid atmospheric changes.
The new model analyses how atmospheric conditions evolve over time rather than treating each observation independently. By identifying relationships between successive measurements of vertical wind movement, it captures both gradual and sudden shifts in the boundary layer while reducing detection delays from hours to just minutes.
Researchers say the improved forecasting system could significantly enhance public health responses during wildfire smoke events by enabling earlier air quality alerts. It could also improve weather forecasting, predict aircraft turbulence, and track the spread of airborne particles such as pollen.
The study demonstrates how artificial intelligence, statistics and atmospheric science can work together to strengthen environmental monitoring and disaster preparedness. Scientists believe the approach could become an important tool for improving air quality forecasting as wildfire smoke events become more frequent and widespread.